{ "$schema": "https://pointcast.xyz/for-agents", "id": "pointcast:next-models:2026-08-31", "title": "Next Models", "description": "A sourced PointCast field guide to the model horizon: controlled frontier systems, Chinese open weights, and the new moving-image stack.", "site": "https://pointcast.xyz", "humanUrl": "https://pointcast.xyz/next-models", "self": "https://pointcast.xyz/next-models.json", "researchAsOf": "2026-08-31", "generatedAt": "2026-09-02T19:20:56.368Z", "modelCount": 9, "vocabulary": { "lane": { "frontier": "Controlled frontier", "open-weights": "Open-weight wave", "moving-image": "Moving-image systems" }, "releaseState": { "upcoming": "Officially acknowledged but not publicly released.", "controlled": "Released only to a limited or vetted audience.", "available": "Available through at least one first-party product or API.", "open-weights": "Model weights are downloadable; inspect each model license before use." } }, "editorialPolicy": { "primarySourcesFirst": true, "vendorBenchmarksAreVendorClaims": true, "rumorsExcluded": true, "terminology": "Open-weight is used unless a release includes enough of the full training stack to support a stronger open-source claim." }, "thesis": [ "Access policy is now a model feature.", "Open weights are a deployment spectrum, not a synonym for open source.", "Video is becoming an editing system, not just a prompt-to-clip trick." ], "lanes": [ { "id": "frontier", "label": "Controlled frontier", "short": "Frontier", "thesis": "The release boundary is becoming part of the model: capability, access and safeguards now ship together.", "modelIds": [ "openai-astra", "anthropic-mythos-5" ] }, { "id": "open-weights", "label": "Open-weight wave", "short": "Open weights", "thesis": "The center of gravity is multipolar. The practical question is no longer whether open weights matter, but which license and serving footprint fit the job.", "modelIds": [ "kimi-k3", "glm-5-2", "qwen-3-8-27b", "deepseek-v4-flash" ] }, { "id": "moving-image", "label": "Moving-image systems", "short": "Video", "thesis": "Video models are turning into multimodal editing systems: references, audio, continuity and revision matter more than a single spectacular clip.", "modelIds": [ "seedance-2-5", "kling-3", "veo-3-1" ] } ], "models": [ { "id": "openai-astra", "name": "Astra", "maker": "OpenAI", "lane": "frontier", "releaseState": "upcoming", "released": "Not released", "access": "Unavailable; external safety testing planned", "license": "Proprietary", "modalities": [ "language", "code", "tools" ], "signal": "A model whose deployment clock is being set by cyber-capability evaluations, not only product readiness.", "humanRead": "The headline is not a benchmark. OpenAI says preliminary evidence may place Astra at its Critical cybersecurity threshold and has slowed work while it strengthens safeguards.", "agentRead": "No public model ID, API surface, price, context window or stable capability spec. Do not treat GPT-6 naming, parameter counts or launch dates as confirmed.", "watchFor": "A system card, public model identifier, access policy and the mitigations attached to any release.", "facts": [ "Officially described as an upcoming model", "Critical cyber capability cannot yet be ruled out", "Not the model involved in the Hugging Face incident" ], "sources": [ { "label": "OpenAI · cyber frontier response", "url": "https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/", "published": "2026-08-07" }, { "label": "OpenAI · pacing model development", "url": "https://openai.com/index/pacing-model-development-cyber-capabilities/", "published": "2026-08-18" } ] }, { "id": "anthropic-mythos-5", "name": "Claude Mythos 5", "maker": "Anthropic", "lane": "frontier", "releaseState": "controlled", "released": "2026-06-09", "access": "Vetted partners; broader access not promised", "license": "Proprietary", "modalities": [ "language", "code", "cyber", "biology" ], "signal": "One underlying model, two deployment envelopes: Mythos 5 for vetted partners and Fable 5 with stronger safeguards for general use.", "humanRead": "Mythos is the clearest example of capability-gated distribution becoming a product tier. The same underlying model can reach different audiences with different safeguards.", "agentRead": "Treat Mythos 5 and Fable 5 as access variants, not interchangeable API aliases. Mythos availability is narrow and organization-vetted.", "watchFor": "Whether restricted capabilities graduate into broad access—or remain a permanent high-trust tier.", "facts": [ "Same underlying model as Claude Fable 5", "Focused on cybersecurity and biology research", "Distributed through Project Glasswing partners" ], "sources": [ { "label": "Anthropic · Claude Mythos 5", "url": "https://www.anthropic.com/claude/mythos", "published": "2026-06-09" }, { "label": "Anthropic · Fable 5 and Mythos 5", "url": "https://www.anthropic.com/news/claude-fable-5-mythos-5", "published": "2026-06-02" }, { "label": "Anthropic · system cards", "url": "https://www.anthropic.com/system-cards" } ] }, { "id": "kimi-k3", "name": "Kimi K3", "maker": "Moonshot AI", "lane": "open-weights", "releaseState": "open-weights", "released": "2026-07-27 weights", "access": "Weights + API + consumer apps", "license": "Kimi K3 custom license", "context": "1M tokens", "modalities": [ "text", "image", "video understanding", "tools" ], "signal": "The first open 3T-class model: 2.8T total parameters, native vision and extremely sparse expert activation.", "humanRead": "K3 makes “frontier open weights” literal at enormous scale. The catch is equally important: open-weight does not mean lightweight, and its custom license deserves a real read.", "agentRead": "2.8T MoE; 16 of 896 routed experts per token; 1M context. Budget for multi-node inference and verify Kimi K3 License obligations before deployment.", "watchFor": "Independent serving reports, quantization quality and how often teams use the weights rather than a hosted API.", "facts": [ "2.8T total parameters", "Native visual understanding", "Weights and technical report released together" ], "sources": [ { "label": "Kimi · K3 technical blog", "url": "https://www.kimi.com/en/blog/kimi-k3", "published": "2026-07-17" }, { "label": "Moonshot AI · Kimi K3 repository", "url": "https://github.com/MoonshotAI/Kimi-K3", "published": "2026-07-27" } ] }, { "id": "glm-5-2", "name": "GLM-5.2", "maker": "Z.ai", "lane": "open-weights", "releaseState": "open-weights", "released": "2026-06-17", "access": "Weights + API", "license": "MIT", "context": "1M tokens", "modalities": [ "text", "code", "tools" ], "signal": "A permissively licensed long-horizon agent model with a full million-token context.", "humanRead": "GLM-5.2 is the cleanest “use it and modify it” proposition in this group: a strong coding and agent model, MIT-licensed, with hosted and self-served paths.", "agentRead": "753B-parameter MoE on the published model card; selectable thinking effort; official local serving support includes Transformers, vLLM and SGLang.", "watchFor": "Reliability across truly long tasks, not just maximum context acceptance.", "facts": [ "MIT-licensed weights", "1M-token context", "Built for long-horizon coding and tool use" ], "sources": [ { "label": "Z.ai · GLM-5.2 release", "url": "https://z.ai/blog/glm-5.2", "published": "2026-06-17" }, { "label": "Z.ai · GLM-5.2 model card", "url": "https://huggingface.co/zai-org/GLM-5.2" } ] }, { "id": "qwen-3-8-27b", "name": "Qwen3.8-27B", "maker": "Alibaba · Qwen", "lane": "open-weights", "releaseState": "open-weights", "released": "2026-08-14", "access": "Weights + API + local runtimes", "license": "Apache 2.0", "context": "262K tokens", "modalities": [ "text", "image", "code", "tools" ], "signal": "A current-generation, dense multimodal agent model at a size teams can plausibly own.", "humanRead": "The 27B release may be more consequential than the 2.4T flagship: it packages the Qwen3.8 generation into a far more deployable footprint under Apache 2.0.", "agentRead": "Dense 27B; 262,144-token serving examples; OpenAI-compatible local routes documented for Transformers, vLLM and SGLang.", "watchFor": "High-quality MLX/GGUF quantizations and real memory/latency reports on workstation-class hardware.", "facts": [ "Dense 27B model", "Apache 2.0", "Native image-text input" ], "sources": [ { "label": "Qwen · Qwen3.8 repository", "url": "https://github.com/QwenLM/Qwen3.8", "published": "2026-08-14" }, { "label": "Qwen · 27B model card", "url": "https://huggingface.co/Qwen/Qwen3.8-27B", "published": "2026-08-14" } ] }, { "id": "deepseek-v4-flash", "name": "DeepSeek V4 Flash", "maker": "DeepSeek", "lane": "open-weights", "releaseState": "open-weights", "released": "2026-07-31 update", "access": "Weights + API", "license": "MIT", "context": "1M tokens", "modalities": [ "text", "code", "tools" ], "signal": "A 284B/13B-active model built to make million-token work cheaper—and improved through post-training rather than another architecture change.", "humanRead": "Flash is the efficiency story: a large model that activates a small fraction of its weights and ships with a permissive license. The July update concentrated on agent behavior.", "agentRead": "284B total / 13B active in the base release; 1M context; MIT. The 0731 update keeps the architecture and changes post-training.", "watchFor": "Whether the agent gains survive across harnesses and production tool environments.", "facts": [ "284B total / 13B active", "MIT-licensed weights", "July update was post-training only" ], "sources": [ { "label": "DeepSeek · V4 release", "url": "https://deepseek.com/en/news/v4-preview/", "published": "2026-04-24" }, { "label": "DeepSeek · V4 Flash 0731", "url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "published": "2026-07-31" }, { "label": "DeepSeek · API changelog", "url": "https://api-docs.deepseek.com/updates/" } ] }, { "id": "seedance-2-5", "name": "Seedance 2.5", "maker": "ByteDance · Dreamina", "lane": "moving-image", "releaseState": "available", "released": "2026-07-31", "access": "Dreamina; BytePlus rollout", "license": "Hosted product", "modalities": [ "text", "image", "video", "audio" ], "signal": "Thirty-second continuous generation, up to fifty references and targeted editing in one system.", "humanRead": "Seedance is pushing the medium away from one-shot clips and toward an edit loop. Longer continuity and many references matter because they reduce the stitching and reroll tax.", "agentRead": "Hosted multimodal workflow. Availability, feature caps and API access can vary by product and region; resolve current account/API terms before automating.", "watchFor": "How much of the advertised 30-second / 50-reference envelope is consistently available outside first-party creative tools.", "facts": [ "Up to 30-second continuous clips", "Up to 50 multimodal references", "Targeted video editing" ], "sources": [ { "label": "Dreamina · Seedance 2.5 launch", "url": "https://dreamina.capcut.com/resource/seedance-2-5-launch", "published": "2026-07-31" }, { "label": "BytePlus · model catalog", "url": "https://ai.byteplus.com/en/model", "published": "2026-08-07" }, { "label": "ByteDance Seed · Seedance 2.0 report", "url": "https://seed.bytedance.com/en/seedance2_0", "published": "2026-02-12" } ] }, { "id": "kling-3", "name": "Kling AI 3.0", "maker": "Kuaishou", "lane": "moving-image", "releaseState": "available", "released": "2026-02-05", "access": "Kling AI product and partner surfaces", "license": "Hosted product", "modalities": [ "text", "image", "video", "audio" ], "signal": "A unified generation-and-editing system with native multilingual audio and multi-shot control.", "humanRead": "Kling 3.0 competes on direction, not only rendering: references, shots, sound and edits live in the same workflow, with clips up to fifteen seconds.", "agentRead": "Hosted product; 3.0 family includes Video, Video Omni, Image and Image Omni. Confirm the exact variant behind any third-party endpoint.", "watchFor": "Native 4K availability, identity consistency and whether Omni features become predictable API primitives.", "facts": [ "Up to 15-second clips", "Native multilingual audio", "Text/image/audio/video workflow" ], "sources": [ { "label": "Kuaishou · Kling AI 3.0 launch", "url": "https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-launches-30-model-ushering-era-where-everyone-can-be", "published": "2026-02-05" }, { "label": "Kuaishou · Q2 2026 update", "url": "https://kuaishou.gcs-web.com/news-releases/news-release-details/kuaishou-technology-announces-second-quarter-and-interim-2026", "published": "2026-08-19" } ] }, { "id": "veo-3-1", "name": "Veo 3.1", "maker": "Google DeepMind", "lane": "moving-image", "releaseState": "available", "released": "Updated 2026-01-13", "access": "Gemini, Flow and Google Cloud", "license": "Hosted product", "modalities": [ "text", "image", "video", "audio" ], "signal": "A mature production path where audio, reference “ingredients,” vertical output and 4K upscaling are first-class controls.", "humanRead": "Veo’s advantage is the surrounding filmmaking surface: consistent ingredients, sound and delivery formats are turning generation into a repeatable production tool.", "agentRead": "Hosted APIs and product surfaces differ. Current published material describes six-second generations with 1080p/4K output paths and native audio.", "watchFor": "Longer native duration, stable character continuity and which Flow controls reach the API.", "facts": [ "Native audio", "Reference ingredient controls", "1080p and 4K output paths" ], "sources": [ { "label": "Google DeepMind · Veo", "url": "https://deepmind.google/models/veo/" }, { "label": "Google · Veo 3.1 Ingredients to Video", "url": "https://blog.google/innovation-and-ai/technology/ai/veo-3-1-ingredients-to-video/", "published": "2026-01-13" } ] } ], "agentNotes": [ "Prefer model.id as the stable key; names and product packaging may change.", "Before production use, re-check the linked primary source for access, license, pricing and model identifiers.", "Do not infer unreleased Astra specifications from this packet. Missing fields are intentionally unknown.", "For hosted video systems, confirm which first-party model variant a provider endpoint actually serves." ], "related": { "currentStack": "https://pointcast.xyz/ai-stack.json", "agentManifest": "https://pointcast.xyz/agents.json", "humanAgentGuide": "https://pointcast.xyz/for-agents" } }